Dialogue Cut

Dialogue Cut creates reviewable dialogue-only cuts from local video files. It can use embedded subtitles when they are useful, or generate German timestamps with MLX Whisper, then render a QuickTime-safe H.264/AAC MP4.

It also includes a video-processing tab for slowing audio and video together, which is useful when preparing footage for dubbing practice.

The desktop app also has a material-grabber tab that wraps yt-dlp for this workflow: paste one URL, click Start, then choose video quality and exact subtitle tracks before the same button changes to Download.

Use it only with media you have the right to process.

Desktop App

cd ui
npm install
npm run tauri dev

CLI

python3 dialogue-only.py /path/to/movie.mkv

Force a fresh German Whisper transcript:

python3 dialogue-only.py /path/to/movie.mkv --force-transcribe

Tune the cut feel:

python3 dialogue-only.py /path/to/movie.mkv --pre-pad 0.3 --post-pad 0.5 --merge-gap 1.0

Slow-Down Transcode

In the desktop app, open Video processing, choose a video, set the playback speed, and start the transcode. A 0.50x output is twice as long as the input. Audio is slowed with ffmpeg atempo, so pitch is preserved.

The tab also shows a thumbnail gallery of videos from the Material grabber Output settings folder. Click a thumbnail to select that video for slow-down.

The output is written beside the source video:

movie.slow-0.50x.mp4

Converter

In the desktop app, open Converter, choose a video (typically an MKV), and start the conversion. The MP4 is written beside the source with the same name (or movie.converted.mp4 if that name is taken). Two modes:

• Re-encode H.264/AAC (default): most compatible for editing; re-encodes everything, so it takes a while.
• Fast remux: copies the video stream as-is into MP4 and converts audio to AAC. Nearly instant, but editors may struggle with HEVC/AV1 sources.

Embedded subtitles and chapters are dropped; only the first video and audio streams are kept.

Transcription

In the desktop app, open Transcription, choose a video or audio file, pick the spoken language (German by default, or auto-detect), and tick the output files you want. The app extracts the first audio track, transcribes it with MLX Whisper, and writes the selected files beside the source:

• movie.de.srt / movie.de.vtt — subtitles (on by default)
• movie.de.txt — plain text transcript (on by default)
• movie.de.json — raw Whisper output with segments and timestamps
• movie.de.tsv — start/end/text timestamp table

With auto-detect the language suffix is omitted.

Material Grabber

In the desktop app, open Material grabber, paste a URL, and click Start. The app asks yt-dlp for metadata first, then shows available video heights and exact subtitle language tracks. The output folder defaults to ~/Downloads/Dialogue Cut Material; change it from Output settings only when you want a different destination. Each URL is saved into its own Title [id] subfolder. The same button changes to Download after metadata is loaded. yt-dlp is not bundled with the app: the managed copy downloads on the first scan. The Grabber checks it against the latest PyPI release and shows an Update button when an older version must be replaced. For YouTube, the managed bgutil provider performs playback verification through the private Deno runtime without opening a browser. The app uses a fixed yt-dlp preset:

• one URL at a time, no playlist expansion
• selected quality cap, best available, or No video
• MP4 remux when possible
• optional selected manual or generated subtitles converted to SRT
• optional raw automatic captions as JSON3 for the selected subtitle languages
• exact subtitle language keys from metadata, not wildcard language matching

Distribution

Build an Apple Silicon macOS app and DMG:

cd ui
npm run bundle:mac

The packaged app does not bundle its large and frequently updated runtime dependencies. It installs a checksum-verified private Python runtime, static ffmpeg/ffprobe, MLX Whisper dependencies, and yt-dlp only when needed. Whisper model files download automatically on first transcription and stay cached in the app's Application Support directory. The managed yt-dlp copy can be updated independently from the app in Material Grabber.

Current packaged target: Apple Silicon Mac with macOS 13.5 or newer.

Solo

The repo includes a Solo command:

Dialogue Cut UI

It runs npm run tauri dev from ui.

Details

See DIALOGUE_ONLY.md for the full workflow, project JSON format, and manual adjustment notes.